• wewbull@feddit.uk
    link
    fedilink
    English
    arrow-up
    17
    ·
    2 months ago

    An LLMs “intent” is always to give you a plausible response even if it doesn’t have the “knowledge”. The same behaviour in a human would be classed as lying IMHO.

    • ContrarianTrail@lemm.ee
      link
      fedilink
      English
      arrow-up
      5
      arrow-down
      3
      ·
      2 months ago

      But you wouldn’t call it lying if a person tells you something they think is true but turns out to be false. Lying means intentionally giving out false information. LLMs don’t have intentions.

      • ℍ𝕂-𝟞𝟝@sopuli.xyz
        link
        fedilink
        English
        arrow-up
        6
        arrow-down
        1
        ·
        2 months ago

        Yeah I think it’s more fitting to use the term bullshitting.

        LLMs actually know that some of their answers have low probability to be the right ones, they give them out regardless, and don’t mention the low confidence of it.

        • ContrarianTrail@lemm.ee
          link
          fedilink
          English
          arrow-up
          1
          arrow-down
          1
          ·
          2 months ago

          Depends which definition of bullshit you use, I guess.

          Frankfurt determines that bullshit is speech intended to persuade without regard for truth. The liar cares about the truth and attempts to hide it; the bullshitter doesn’t care whether what they say is true or false.

          Wiki

          • conciselyverbose@sh.itjust.works
            link
            fedilink
            English
            arrow-up
            2
            ·
            2 months ago

            the bullshitter doesn’t care whether what they say is true or false.

            That’s another way to say “intent is irrelevant”.

            It’s also effectively the perfect definition of LLM output. Content for the sole purpose of looking the part with absolutely no consideration for reality.

            • ContrarianTrail@lemm.ee
              link
              fedilink
              English
              arrow-up
              1
              arrow-down
              2
              ·
              2 months ago

              …bullshit is speech intended to persuade…

              Quoting out of context is not going to score you any points

              • conciselyverbose@sh.itjust.works
                link
                fedilink
                English
                arrow-up
                1
                ·
                2 months ago

                It is intended (by the designer) to persuade. It’s intended to persuade you that it’s something a human would say.

                Ignoring that you’re trying to claim one dude’s definition of bullshit as the law, that one dude’s definition is an exact flawless match for what LLMs are.

                • ContrarianTrail@lemm.ee
                  link
                  fedilink
                  English
                  arrow-up
                  1
                  arrow-down
                  1
                  ·
                  2 months ago

                  It is intended (by the designer) to persuade.

                  According to you, I presume? Or can you back that up somehow?

                  LLMs were developed to simulate human-like understanding and generation of language. They’re called large language models for a reason.

      • wewbull@feddit.uk
        link
        fedilink
        English
        arrow-up
        1
        ·
        2 months ago

        …but if they don’t know I expect them to say so. An LLM isn’t trustworthy until it says “I don’t know”.